A few years ago I did a similar comparison for older Olympus models, to let camera users compare between various cards on the same camera, or various cameras using the same card. If the speed increase offered by a newer card will show on a given camera, it may be worth the price premium over an older one.

At that time, the cameras were much slower than today: a typical writing time for a raw image file was 10 seconds or more. Therefore I could get away with the simplest possible measurement technique: taking a picture and then using a stopwatch to measure the time during which the cameras red "busy" light stays on. At these times, error introduced by the differences in my reaction time was small enough: after all, 1/4 of a second (my estimate) is just 2.5% of the measured value, an insignificant value.

Then the cameras became faster. Modern digital SLR can write a raw file is less than a second. For this, 0.25 s is a 25% error or more — a significant difference, and as the cameras become faster, it becomes even worse.

Thus, instead of maintaining the old article by adding new cameras and card types, I had no choice but to start from scratch, with this article as a result.

Measurement method

To eliminate any statistical and/or systematic errors due to variations in human reaction speed, the method is based on photographing the face of a running stopwatch, with the camera shooting in the sequential mode. The number of frames has to be large enough for the camera to run out of buffer memory; usually it is best just to keep shooting for 30 or 40 seconds.

Then the images are reviewed, and each frame number is assigned a time value as per the stopwatch reading shown in it. (This really does not have to be done for each frame, see below.)

The first part corresponds to the camera writing into its fast memory buffer; for the E-30 the slope is 0.2 s per frame; its reciprocal is the burst frame rate of that camera: 5 frames/s (FPS).

When the buffer becomes full, the slope will be defined by how fast the camera writes the buffer contents to the card memory. Therefore the slope after the bend is actually the card writing time per frame, what we want to measure.

There is no need to use all frames from your series. It is fully enough and equally accurate to capture the time of one of the first frames after the slow-down, t_slow (here t₁₅ will do just fine for both cameras, even if for the E-510 any frame after the first ten would be fine), and then the last one. The card writing time can be then computed as Δt = (t_last-t_slow)/(N_last-N_slow). All you need is to capture two times and their frame numbers.

Measurement accuracy (maximum error) will be dt/(N_last-N_slow), where dt is the precision of the stopwatch used (here: 0.1 s). Therefore ten frames after the slowdown is already enough to assure that the error of the result stays within 0.01 s.

Caveats of this method are few, but they have to be remembered:

• It will work only for cameras with shooting rate higher than the card writing one. This is so far the case, but quite soon we may see cards matching anything the camera will throw at them. Then our method will measure just the time needed by the camera to perform all operations other than writing, not the writing time.
• The camera has to be capable of writing to the card simultaneously to (and with a speed not affected by) any other functions performed at the moment, including (but not limited to) operating the shutter, providing the monitor display, reading the signal off the sensor, and processing it. This seems to be the case in current SLRs by Olympus and other makers, but not necessarily for all other cameras. If the assumption is not met, the card writing times will be overestimated.
• If the results are expressed per frame (as opposed to per megabyte), they will depend on the size of files being written, even for the same camera. This is because some cameras use compression for raw images (in this comparison: all except for the E-500), and some images (depending on the amount of captured detail, and ISO setting) may compress better than others.

  For a given camera and the same subject of test pictures, variations in file size are not significant. (This is why I am arranging all series in the same way: the watch face filling the frame vertically, uniform black background). For example, file sizes in various series of my E-30 measurements, shot on different days under different light, ranged from 11.3 to 11.6 MB — a quite narrow spread, with different frames within the same series staying the same (when rounded to the nearest 0.1 MB).

I also set the camera to manual focus and metered manual exposure (prefocusing before the first frame), and ISO to 400.

Do you need a stopwatch? Not necessarily. Using the same method with a regular watch (preferably analog), with the readout precision of one second, you may still get the write times with an accuracy of 0.02 s or so (a 5% error for the fastest cards of 2008) if you collect at least fifty frames in the slow part of the sequence. Or, without a watch at all, you can rely on the DateTimeOriginal field in the EXIF data.

Results

Here are the results: the time it takes for a single raw image to be written out to a card, for a number of cards I happen to have, and a number of cameras, all of them Olympus SLRs. (The E-410 results should be identical to those for the E-510; for E-420 and E-520 — the same or slightly faster).

While the measured time values are accurate up to 0.01 s (0.02 s dor xD-Picture), they are affected to a larger degree by raw file sizes (see Caveat 3 above).

The data for the E-500, shown for reference, were derived using a different technique. That camera would stop the high-speed ORF sequence after four frames, so I had to capture manually the time between the first shutter activation and the end of writing. Assuming a manual timing error within 0.5 s, this would mean a measurement accuracy of 0.1 s or so.

Average file sizes for my stopwatch images are also shown, so that you can convert my data to MB/s if needed.

Comments

Raw image files in the E-30 are 10-20% larger than in ten-megapixel E-3 and this is reflected in writing times. An exception is when the camera is used with a really fast card (here: the 8 GB Extreme IV); this means the E-30 takes advantage of the SanDisk writing acceleration technology. (This may or may not be the case with similar technologies used by other manufacturers.)

The Lexar Professional UDMA (with a 300× speed rating) is 20% slower than the Extreme IV in a fast camera (E-30).

The E-620, in spite of being a slightly newer model than the E-30, is not quite as fast in writing with the fastest SunDisk card.

The 8 GB Extreme IV can be, with a fast camera, almost twice as fast as the 2 GB version. Definitely, it uses another technology. I am not sure whether this is because of greater capacity or just more recent vintage.

Transcend 266× is based on single-level storage technology. The same brand offers similar but multi-level cards in the budget category; these I would expect to be twice slower.

xD-Picture M+ cards (the only kind currently manufactured) take 20-25% longer to write than the "H" type from three years ago. Generally, xD cards are no competition to almost any Compact Flash ones (speed- or otherwise); see here.